Gemma, DeepMind's Family of Open Models
summary
Gemma 4 is a released family of open models ranging 2B–32B parameters that runs on Android and iPhone and even Raspberry Pi with demos of ten parallel instances generating SVGs on-device using llama cpp The architecture introduces E2B per-layer embeddings to offload parameters to CPU or disk for low latency inference The release moves to an Apache 2 license and reports 10 million downloads for Gemma 4 base models and 500 million for the family plus validated lab results for cancer therapy pathway suggestions
AI-written summary · about
chapters (15)
- 0:00Introduction to the Gemma model family
- 0:41Evolution from Gemma 3 to Gemma 4
- 1:21Overview of the new Gemma 4 capabilities
- 2:31Live demonstrations of on-device applications
- 3:38LM Arena scores and performance benchmarks
- 5:07Apache 2 license transition
- 5:27Technical deep dive: The E2B architecture and per-layer embeddings
- 6:57Multimodal understanding and multilingual support
- 8:43Ecosystem growth and community adoption
- 10:07Product integrations, including Android Studio
- 10:46Statistics on model downloads and fine-tuning
- 11:27Official Gemma variants: Shield Gemma and MedGemma
- 12:16Community research and sovereign AI efforts
- 12:56Real-world applications, from cancer therapy to offline tasks
- 14:05Closing remarks and future outlook
related talks
10mAI for everyone with Gemma
16mOpen models in the Gemini era
40mBuilding amazing Front-End Project with Gemma: A State-of-the-Art Open Model
44mBuilding amazing Front-End Project with Gemma: A State-of-the-Art Open Model
39mFlutter Project with Gemma: A State-of-the-Art Open Model
10mDemo: Using Gemma with the Hugging Face ecosystem
12mWelcome to Gemma 3
14mGemma on mobile and web. Best and worst practices